def test_concat_layer_wrapper(inputs, concat_stats, mode, clip_acts,
expected_output):
with pytest.raises(ValueError):
# Check exception on wrong layer type
RangeLinearQuantConcatWrapper(torch.nn.Module(), 8, mode, clip_acts,
concat_stats)
layer = distiller.modules.Concat(dim=1)
with pytest.raises(NotImplementedError):
# Check exception on no stats
RangeLinearQuantConcatWrapper(layer,
8,
mode,
clip_acts,
activation_stats=None)
for idx in range(len(inputs)):
inputs[idx] = attach_quant_metadata(inputs[idx],
8,
mode,
stats=concat_stats['inputs'][idx],
clip_mode=clip_acts,
per_channel=False,
num_stds=None,
scale_approx_mult_bits=None)
model = RangeLinearQuantConcatWrapper(layer, 8, mode, clip_acts,
concat_stats)
model.eval()
output = model(*inputs)
torch.testing.assert_allclose(output, expected_output)
def test_concat_layer_wrapper(inputs, concat_stats, mode, clip_acts, expected_output):
with pytest.raises(ValueError):
# Check exception on wrong layer type
RangeLinearQuantConcatWrapper(torch.nn.Module(), 8, mode, clip_acts, concat_stats)
layer = distiller.modules.Concat(dim=1)
with pytest.raises(ValueError):
# Check exception on no stats
RangeLinearQuantConcatWrapper(layer, 8, mode, clip_acts, activation_stats=None)
model = RangeLinearQuantConcatWrapper(layer, 8, mode, clip_acts, concat_stats)
model.eval()
output = model(*inputs)
torch.testing.assert_allclose(output, expected_output)